At the end of a campaign, when most pollsters are releasing polls, a simple average can be quite accurate in forecasting the election. But in the months leading up to Election Day, polls are more sparse, and a partisan poll or a survey from a pollster with a large house effect can pull the average this way or that — showing a trend that’s probably just the function of when polls happen to have been published.
That’s been a large part of the story in Alaska, where the FiveThirtyEight Senate forecast has Republican Dan Sullivan leading. Sullivan’s lead, in turn, has helped boost Republicans’ overall probability of taking the Senate (nearly 60 percent according to our latest update).
I detailed Tuesday how different polls in the 49th state have shown different things. Independent polls — polls not sponsored by any political organization — had Sullivan up by an average of 3.3 percentage points since the August primary. Polls sponsored by liberal organizations had Democratic Sen. Mark Begich ahead by 3.7 percentage points on average. When all the polls were averaged, Begich led by an average of 0.2 percentage points.
Yet, it would be a mistake to just throw them into an average or median. The sponsored polls show the race leaning in one direction, and the independent polls found the opposite.
Senate polls sponsored by partisan groups have had a bias in favor of the candidate the group wanted to win. That’s why FiveThirtyEight corrects for this house effect, and why we’ve consistently given Sullivan a slight advantage over the past few weeks. On Tuesday, we had him forecasted to win by 2.5 percentage points.
As of Friday, Sullivan is projected to win by about 3 percentage points — a small change even though several new polls showing Sullivan in front have been released. A Rasmussen Reports survey taken this week had Sullivan at 48 percent and Begich at 43 percent. A Dittman Research poll taken last week for the conservative leaning Chamber of Commerce had Sullivan at 49 percent to Begich at 43 percent. And a poll done by Marc Hellenthal over the past two weeks found Sullivan up 46 percent to 42 percent.
Put those all together, and the average poll since the Republican primary has Sullivan up 1.6 percentage points. If we took a median of the polls Tuesday, it would have shown Sullivan up 0.5 points. On Friday, it had Sullivan ahead by 2 points. In both cases, the race moved much more than the FiveThirtyEight forecast.
The reason is rather simple: An unadjusted average or median doesn’t account for the fact that we know different kinds of polls and different pollsters tend to have predictable leans. If all you’re using is a simple average or median to look at an election, the order in which pollsters happen to release polls can make a race look more volatile than it is, or make a candidate look like she’s surging when she isn’t.
Instead, it’s likely the race in Alaska has been fairly stable over the past few weeks. Sullivan is ahead by a couple of percentage points.